Nemotron 3.5 Super VL Ollama: What You Need To Know

To run Nemotron 3.5 Super VL Ollama style, install Ollama, type ollama run nemotron-3.5-super, and wait for a download of roughly 80-90GB. That is the whole setup. The catch is the hardware. This is a 120B model, so it needs a machine with serious memory, not a normal laptop.

📺 Watch: NVIDIA Nemotron 3.5 Super VL Is INSANE! (FREE AI Model)

🔥 Get the Agent OS as a free bonus: AI Profit Boardroom members get the full Agent OS zip, prompt libraries, daily tutorials and weekly live coaching calls. → Get inside

I'll walk you through the exact commands, how to plug it into Hermes, OpenCode, Claude Code and OpenClaw with one line each, and what to do if your machine can't handle it. Short version on that last part: most people should use it hosted and skip the local install entirely.

If you want the whole setup done for you, my AI Profit Boardroom ($69/mo) includes the Agent OS zip, a video tutorial and a 30-day roadmap, so you can swap new open models like this one in without guessing.

Why Ollama Calls It nemotron-3.5-super

This trips people up. You search Ollama for "Super VL" and you see nemotron-3.5-super instead. Same model. The Ollama library name is nemotron-3.5-super, and Ollama describes it as "NVIDIA's vision-language model for coding, reasoning, tool use, and document understanding."

So when you see "Nemotron 3.5 Super" in Ollama, that is the vision model. It takes text and images in and gives text out. You are not getting a cut-down text-only version.

Here's what the Ollama listing shows:

Name: nemotron-3.5-super
Size: 120b, roughly 80-90GB on disk
Context window: 1M tokens
Tags: vision, tools, thinking
Input: text and images

Those three tags matter. Vision means you can hand it a screenshot or a document. Tools means agent harnesses can let it call functions. Thinking means it supports a reasoning mode you can switch on or off. If you want the full background on what the model actually is, read my Nemotron 3.5 Super VL overview.

How to Set Up Nemotron 3.5 Super VL Ollama in 3 Steps

Step 1: Install Ollama

Go to the Ollama website, download the app for Mac, Windows or Linux, and install it like any other program. Open your terminal once it's done so you can type commands.

Step 2: Pull and run the model

Type this and press enter:

ollama run nemotron-3.5-super

Ollama downloads the model the first time. At roughly 80-90GB, that takes a while even on fast internet. Once it's done, you get a chat prompt right in the terminal. Ask it something to check it works.

Step 3: Plug it into your agent

This is the part I like. Ollama has launch commands that open an agent harness with the model already selected. One line each:

Hermes: ollama launch hermes --model nemotron-3.5-super
OpenCode: ollama launch opencode --model nemotron-3.5-super
Claude Code: ollama launch claude --model nemotron-3.5-super
OpenClaw: ollama launch openclaw --model nemotron-3.5-super

Pick the harness you already use. If you don't use any yet, OpenCode is the easiest place to start. My Agent OS guide shows how I run Hermes, OpenClaw, OpenCode and Claude side by side and swap models in and out.

📺 Watch: OpenClaw + Nvidia Nemotron 3 Super + Ollama is INSANE!

The Hardware Reality

Let me be straight with you. This is not a normal-laptop model.

The model has 120B total parameters. It's a mixture-of-experts design with 12B active per token, which helps speed, but all the weights still have to sit in memory. That's why the Ollama download is roughly 80-90GB, and you need about that much memory free to run it, plus extra for images and long context.

Machines that can realistically run it locally:

A 128GB NVIDIA DGX Spark
A big-memory workstation with 128GB or more
High-memory Apple machines with enough unified memory to hold the model plus your working space

Machines that won't:

A typical 16GB or 32GB laptop
Most gaming PCs with a single consumer GPU
Anything where 80-90GB of free memory isn't realistic

And the 1M context window is a ceiling, not a promise. Usable context depends on your memory and runtime. Every extra chunk of context eats more memory, so a machine that loads the model fine can still choke on a giant prompt.

If you're weighing up the extra memory a home setup needs, my LM Studio setup for Nemotron 3.5 Super VL covers the smaller Q4_K_M build at around 70GB.

📺 Watch: New FREE Nvidia Nemotron 3 Super + Ollama

What I Actually Use It For

I've used Nemotron 3.5 Super VL free inside OpenCode, inside my Agentic Operating System. I picked "Nemotron 3.5 Super VL" in the model drop-down and asked it to build a beautiful to-do list app, make it entertaining, engaging. The build came out working. It also shows up as a model I can pick inside Hermes Agent.

My honest take: it's not a frontier model, and it's not going to replace Opus 5.5. But it's free, open, fast, and it can read screenshots, PDFs, charts and receipts. So I slot it in for jobs that start with a picture or a file.

Good jobs for it:

Turning a screenshot of a design into working code
Pulling numbers out of a chart or a PDF report
Reading receipts and invoices into a list
Agent tasks where it needs to call tools

If you want help working out which model fits which job in your business, book a free AI strategy session and we'll map it out.

Speed and Benchmarks, With a Caveat

NVIDIA reports up to 5x higher single-GPU throughput on agentic tasks versus open models its size, helped by multi-token prediction, where an internal drafter proposes several tokens that get checked in one pass. Per NVIDIA, it scores 63.44% on Hermes Bench versus 56.33% for Nemotron 3 Ultra.

Those are NVIDIA's own numbers. No independent replication exists yet. And they're measured on NVIDIA's hardware, not your laptop. Your local speed through Ollama will depend completely on your machine. For how I test models myself, check Goldie Bench.

Troubleshooting Basics

"Not enough memory" or the model won't load

Your machine can't hold the model. Don't fight it. Use it hosted instead. It's on OpenRouter as nvidia/nemotron-3.5-super-vl, including a :free variant, and you can pick it free inside OpenCode. You get the same model without the 80-90GB download.

It loads but runs painfully slow

Usually the model is spilling out of fast memory. Close other heavy apps, keep prompts shorter, and avoid stuffing huge context in. If it's still crawling, hosted is the better call.

The download keeps failing

Check you've got enough free disk space for roughly 80-90GB, plus some headroom. Then run ollama run nemotron-3.5-super again. Ollama picks up where it left off.

The launch command doesn't open my agent

Make sure the harness itself is installed first, and check you typed the model name exactly: nemotron-3.5-super. Not "super-vl". That's the most common mistake.

I ran out of free tokens in OpenCode

Switch to another free model in OpenCode. I've used "Big Pickle". Or run it in Hermes instead.

If you're still on the fence about the base Super model versus this vision one, my breakdown of Nemotron 3.5 Super clears up the naming.

Should You Run It Locally at All?

Here's my rule. If you already own a 128GB DGX Spark or a big-memory workstation, run it in Ollama. You get privacy, no rate limits and the one-line agent launches.

If you don't, use it hosted. Buying hardware just to run one free model makes no sense for most people. OpenRouter's free tier and OpenCode free get you working today.

Inside the AI Profit Boardroom I run 4 weekly coaching calls and post daily tutorials, and 3,400+ members are building with setups like this. If you'd rather talk it through one on one, grab a free AI strategy session.

FAQ

Is nemotron-3.5-super in Ollama the same as Nemotron 3.5 Super VL?

Yes. Ollama lists the vision-language model under the name nemotron-3.5-super, with vision, tools and thinking tags and text plus image input.

How big is the Ollama download?

Roughly 80-90GB. You need about that much memory to run it, plus extra for images and context.

Can I run it on my laptop?

Almost certainly not, unless your laptop has a huge amount of memory. A typical 16GB or 32GB machine can't hold it. Use OpenRouter or OpenCode free instead.